v0.5.1 — 21 providers, 219 models, stateless HTTP

What's new

• 4 new providers: xAI, Inference.net, SEA-LION, Ollama
• Cloudflare AI expanded: 6 → 14 models (verified against live /v1/models)
• Scaleway expanded: 7 → 10 models (verified against live /v1/models)
• Mistral expanded: 1 → 10 models — Mistral Large (S-tier), Pixtral Large (S-tier), Devstral, Magistral, and more
• Codestral endpoint fix: migrated from deprecated codestral.mistral.ai to api.mistral.ai

Bug fix

• Stateless HTTP: stateless_http=True eliminates "Session not found" errors when the server restarts. No more client restarts needed.

Stats

• 219 models across 21 providers
• 17 providers with working free inference

v0.4.1 — Fix MCP Registry publishing

Patch release to fix MCP Registry validation.

• Add mcp-name tag to README for registry ownership verification
• No functional changes from v0.4.0

v0.4.0 — Public release

First public release of Model Radar.

What's new in 0.4.0

• SQLite persistence and live model fetching from provider APIs
• Web dashboard with SSE + Streamable HTTP
• Host swap instructions and setup workflow for agents
• ask() multi-model consensus with raw response access
• is_free / free_only filtering
• Availability check and ping test utilities
• PyPI + MCP Registry automated publishing

Install

pip install model-radar-mcp

MCP config (Claude Code / Cursor)

{
  "mcpServers": {
    "model-radar": {
      "command": "model-radar",
      "args": ["serve"]
    }
  }
}